LETTERShttps://doi.org/10.1038/s41562-018-0366-8

1Department of Cognitive, Linguistic, and Psychological Sciences, Brown University, Providence, RI, USA. 2Institute for Experimental Psychology, University of Regensburg, Regensburg, Germany. 3Department of Neuroscience, Brown University, Providence, RI, USA. 4Present address: Department of Ophthalmology, School of Medicine, New York University, New York, NY, USA. 5Present address: Department of Cognitive and Psychological Sciences, Graduate School of Informatics, Nagoya University, Furo-cho, Chikusa-ku, Nagoya City, Japan. 6These authors contributed equally: Ji Won Bang, Kazuhisa Shibata, Sebastian M. Frank. *e-mail: takeo_watanabe@brown.edu

After encoding, memory traces are fragile and easily dis-rupted by new learning until they are stabilized through a pro-cess termed consolidation1,2. However, several studies have suggested that consolidation does not make memory traces permanently stable. The results of these studies support the theory that the retrieval of previously consolidated memory, termed reactivation, renders the memory traces labile again and subject to disruption by new learning unless they go through a further consolidation process, termed reconsoli-dation3–8. However, it remains controversial whether reac-tivation and reconsolidation occur at a human behavioural level9–11 and whether consolidation and reconsolidation have common mechanisms12,13. Here, we found that reconsolida-tion does occur after reactivation in visual perceptual learn-ing14–25, a type of skill learning, in humans. Moreover, changes in behavioural performance, as well as in concentrations in the excitatory neurotransmitter glutamate and in the inhibi-tory neurotransmitter GABA (γ -aminobutyric acid), as mea-sured by magnetic resonance spectroscopy, in early visual areas exhibit similar time courses during consolidation and reconsolidation. These results indicate that reconsolidation after reactivation and consolidation in humans share common behavioural and neurochemical mechanisms.

In study 1, we tested behaviourally whether visual perceptual learning (VPL) in humans undergoes reactivation and reconsolida-tion. Specifically, we examined whether reactivation, that is, a few blocks of performing the trained visual task, leads to two dynamic states: a fragile state shortly after reactivation and a reconsolidated state 3.5 hours after reactivation. If reactivation makes the consoli-dated VPL fragile once, then a competing new VPL task, which fol-lows immediately after reactivation, would disrupt or interfere with the reactivated VPL. If the reactivated VPL becomes reconsolidated 3.5 hours after reactivation, the competing new VPL task should not interfere with the reactivated VPL.

We conducted experiments with an orientation detection task (see Methods; Fig. 1a). Two orientations (10° and 70°) were randomly assigned to orientation A (the first trained orientation) and orientation B (the second trained orientation) across participants. Our previous study showed that the learning of this task is consolidated after 3.5 hours24. There were two groups (Fig. 1b): the short-interval group (n = 15) and the long-interval group (n = 15). On day 1, there were 16 blocks of training with orientation A for both groups, so that the encoded VPL could be consolidated24 in both

groups on day 1. The only difference between the procedures with these two groups was the time interval between the offset of reacti-vation and the onset of the following test on orientation B on day 2.

The results of study 1 suggested that the reactivated VPL was fragile immediately after reactivation, but was less fragile 3.5 hours after reactivation (Fig. 1c). A two-way mixed analysis of variance (ANOVA) was applied to performance improvement (%) with fac-tors day (day 2 versus day 3) and group (short-interval versus long-interval groups). Performance improvement on day 2 is defined by: ((threshold on the first day − threshold on the second day)/threshold on the first day) × 100 (see Methods below). Performance improve-ment on day 3 is defined by: ((threshold on the first day − threshold on the third day)/threshold on the first day) × 100. If the fragility of reactivated VPL changes over time, a significant interaction should occur between day and group in the ANOVA. The results indicated a significant day × group interaction (F(1,28) = 6.86, P = 0.014, partial η 2 = 0.197). Further post-hoc tests indicated a significant simple main effect of group at day 3 (F(1,28) = 6.451, P = 0.017, partial η 2 = 0.187, 95% confidence interval of the difference (CI): 2.849–26.601) but not at day 2 (F(1,28) = 0.023, P = 0.881). In addi-tion, post-hoc analyses showed that performance improvements on day 3 were significantly worse than on day 2 for the short-interval group (simple main effect of day: F(1,28) = 4.326, P = 0.047, par-tial η 2 = 0.134, 95% CI: 0.12–15.758), whereas no significant dif-ferences were found between day 3 and day 2 for the long-interval group (simple main effect of day: F(1,28) = 2.638, P = 0.116). See Supplementary Table 1 for the raw threshold values.

These results are in accordance with the hypothesis that, after reactivation, VPL on orientation A becomes fragile so that the fol-lowing competing training on orientation B interferes with the old VPL of orientation A, whereas 3.5 hours later, the reactivated VPL on orientation A has become reconsolidated so that no interfer-ence occurs with the following competing training on orientation B. Although it has remained controversial whether reactivation and reconsolidation processes occur in humans, our results strongly suggest that reactivation and reconsolidation indeed do occur in humans with VPL.

Next, we conducted study 2 to investigate whether underlying neurochemical mechanisms related to consolidation and reconsoli-dation are similar in the human visual cortex. Previous studies have shown that the plasticity of cortical regions is positively correlated with the concentration of glutamate26, a major excitatory neurotrans-mitter, whereas it is negatively correlated with the concentration of

Consolidation and reconsolidation share behavioural and neurochemical mechanismsJi Won Bang1,4,6, Kazuhisa Shibata1,5,6, Sebastian M. Frank1,2,6, Edward G. Walsh3, Mark W. Greenlee2, Takeo Watanabe! !1,3* and Yuka Sasaki1,3

NATURE HUMAN BEHAVIOUR | VOL 2 | JULY 2018 | 507–513 | www.nature.com/nathumbehav 507

LETTERS NATURE HUMAN BEHAVIOUR

GABA, a chief inhibitory neurotransmitter27,28. Moreover, numer-ous studies have indicated that the balance between excitatory and inhibitory signals determines the degree of plasticity29. In particular, our recent study24 demonstrates that a typical consolidation process in VPL is highly correlated with the ratio of the concentrations of glutamate to GABA neurotransmitters (E/I ratio) in human early visual areas24 in strong association with psychophysical results: when VPL was in a plastic and therefore unstable state as indicated by significant interference with new learning immediately after encoding, the E/I ratio in the early visual areas increased. However, within a few hours, the E/I ratio returned to baseline levels, sug-gesting that VPL became stable and was not interfered with by new learning. In short, when a state of VPL is plastic and unstable, the E/I ratio in the early visual areas is greater than baseline, but returns to baseline as VPL becomes consolidated.

In study 2, to address the above-mentioned question whether the neural mechanisms of reactivation and reconsolidation pro-cesses are similar or different from those of post-encoding and con-solidation, we investigated the neurochemical dynamics underlying reconsolidation of VPL using magnetic resonance spectroscopy (MRS). There were two groups: the reactivation and the control groups. We measured and compared the E/I ratios in the early visual areas in the reactivation group (n = 12), in which reactivation of consolidated learning occurred, with those in the control group (n = 12), in which no reactivation occurred (see Fig. 2a). The reac-tivation group performed a total of 19 blocks (3 blocks for test and 16 blocks for training) on orientation A on day 1 to induce learning

of orientation A. On day 2, three blocks on orientation A were con-ducted to reactivate learning of orientation A. In the control group, on day 1, there were three blocks of trials on orientation A, fol-lowed by another three blocks of trials on orientation B; this design assured that no learning of orientation A could occur because of retrograde interference from orientation B with orientation A. The procedure on day 2 in the control group was identical to the reacti-vation group. Importantly, the number of MRS scans as well as the intervals between the MRS measurements and the number of blocks were exactly the same between the two groups on day 2. The differ-ence in the experimental manipulation occurred only on day 1.

As predicted, learning of orientation A occurred in the reacti-vation group, whereas no such learning was found for the control group. A one-way ANOVA with the factor group (reactivation ver-sus control) indicated a significant main effect of group on perfor-mance improvement on day 2 (Fig. 2b; F(1,22) = 10.208, P = 0.004, partial η 2 = 0.317, 95% CI: 12.93–60.81). The performance improve-ment on day 2 was significantly larger than 0 for the reactivation group (one-sample t-test, t(11) = 6.282, P = 5.99 × 10−5, Cohen’s d = 1.81, 95% CI: 23.72–49.32) but not for the control group (one-sample t-test, t(11) = − 0.035, P = 0.973).

Figure 2c shows that the E/I ratio in the early visual areas increased in association with reactivation. If reactivation makes VPL unstable and fragile, the E/I ratio should be enhanced24 immediately after reactivation in the reactivation group but not in the control group. For the reactivation group, the E/I ratio sig-nificantly increased by 11.45 ± 5.40% (mean ± s.e.m.) immediately after reactivation compared with the E/I ratio before reactivation and returned to baseline 3.5 hours later. Conversely, for the control group, no significant changes in the E/I ratio occurred across the three time points of MRS measurements. Confirming these differ-ences between the reactivation and control groups with a two-way mixed ANOVA with factors group (reactivation versus control) and session (0 h versus 3.5 h) indicated that there was a significant interaction between group and session (F(1,22) = 5.027, P = 0.035, partial η 2 = 0.186). In addition, there was a simple main effect of session for the reactivation group (F(1,22) = 5.276, P = 0.032, par-tial η 2 = 0.193, 95% CI: 1.14–22.34) but not for the control group (F(1,22) = 0.763, P = 0.392). As mentioned above, a higher E/I ratio indicates a greater degree of plasticity24. Thus, the present results are in accordance with the hypothesis that reactivation of VPL leads to an increase in both plasticity and the E/I ratio in early visual areas. This increased plasticity and the E/I ratio taper off within a few hours of reconsolidation. This is strong evidence that reactivation and reconsolidation indeed occur in early visual areas at least in VPL. See Supplementary Fig. 1 for normalized GABA and gluta-mate concentrations for both groups.

Are the underlying mechanisms of reactivation and reconsolida-tion in VPL similar to those of post-encoding and consolidation? Figure 2d shows the E/I ratio changes in the early visual areas in the post-encoding stage (n = 12)24. The E/I ratio was found to be sig-nificantly increased by 11.97 ± 3.91% (mean ± s.e.m.) 30 min after the end of training relative to pre-training and tapered off 3.5 hours after training24. These results indicate that the time-course changes in the E/I ratio during reconsolidation after reactivation are similar to those during consolidation after encoding.

Is such similarity in neurochemical plasticity between recon-solidation and consolidation as indexed by the E/I ratios also observed when the concentrations of glutamate and GABA in early visual areas are examined separately? To answer this question, we compared changes of each metabolite from the reactivation group in the present study with the previous MRS data collected dur-ing consolidation24 and tested whether the concentration of each metabolite was significantly different between reconsolidation and consolidation. As one of the assumptions for ANOVA, equality of error variances, was not satisfied (see Methods), we conducted a

Day 1 Day 2 Day 3

0 h

3.5 h

Short

Long Per

form

ance

impr

ovem

ent (

%)

LongShort

*

*0

5

10

15

20

25

30

35

40

2 3Day

a

b c

Time

50 ms500 ms

300 ms 50 ms Response

Fig. 1 | Procedures and results of study 1. a, The orientation detection task. The bullseye fixation point is shown disproportionally larger for illustrative purposes. After the fixation point (500!ms), the Gabor orientation appeared either in the first or in the second of the two intervals, which were separated by a 300 ms blank. Participants reported in which interval the Gabor orientation was presented. b, Experimental procedures of study 1. There were the short-interval group (n!= !15) and the long-interval group (n!= !15). The red-filled boxes represent the three blocks of test on orientation A. The red-hatched boxes show 16 blocks of training on orientation A. The blue-filled boxes represent the three blocks of test on orientation B. The blue-hatched boxes show 16 blocks of training on orientation B. c, The mean (± s.e.m.) performance improvements on orientation A for days!2 and 3 relative to day!1 for both the long-interval and the short-interval groups. *P!< !0.05 indicates the results of post-hoc tests of the ANOVA: a significant simple main effect of group at day!3 (P!= !0.017) and a significant simple main effect of day for the short-interval group (P!= !0.047). See the main text for details of the ANOVA results.

NATURE HUMAN BEHAVIOUR | VOL 2 | JULY 2018 | 507–513 | www.nature.com/nathumbehav508

LETTERSNATURE HUMAN BEHAVIOUR

Mann–Whitney U-test and compared the normalized metabolite (GABA and glutamate) concentrations at each time point (0 h and after 3.5 h) between reconsolidation and consolidation. We did not find any significant differences on either normalized metabolite at any time point between consolidation and reconsolidation (GABA at 0 h: U = 93, P = 0.242; GABA at 3.5 h: U = 76, P = 0.843; glutamate at 0 h: U = 103, P = 0.078; glutamate at 3.5 h: U = 79, P = 0.713; all n = 24, P values are uncorrected). These results suggest that recon-solidation and consolidation processes are associated with similar changes in metabolite concentrations over time.

So far, the results from MRS measurements imply that the neu-ral mechanisms underlying consolidation after post-encoding and reconsolidation after reactivation may be very similar. However, because it has been demonstrated that the measurements of gluta-mate and GABA by MRS are of both intrasynaptic and extrasynaptic origin30–32, a similar time course for the E/I ratio and each metabo-lite does not necessarily demonstrate a similar neural mechanism. Thus, we conducted additional behavioural experiments (study 3) to test whether performance changes are significantly different in a specific time window following post-encoding (consolidation) and reactivation (reconsolidation).

In study 3, there were two groups (Fig. 3a), the consolidation group (n = 15) and the reconsolidation group (n = 14). In the con-solidation group, participants trained on orientation A for 16 blocks following a pre-test session (3 blocks), then performed a retest ses-sion 3.5 hours after the offset of training. All measurements were

completed on day 1 for the consolidation group. Participants in the reactivation group performed a pre-test session (3 blocks), which was followed by a training session (16 blocks) on day 1. On the next day (day 2), participants performed a reactivation session (3 blocks) and a retest session (3 blocks) 3.5 hours after the offset of reactivation.

In the consolidation group, VPL was expected to consolidate during the 3.5-hour interval following training on day 1. In the reconsolidation group, reactivated VPL was expected to reconsoli-date during the 3.5-hour interval following reactivation on day 2. If there are different neural mechanisms underlying consolidation and reconsolidation processes, then the performance changes over the 3.5-hour interval should be different between the consolidation and the reconsolidation groups. However, the performance changes (see Methods) after the 3.5-hour interval were not significantly dif-ferent between the consolidation and the reconsolidation groups (paired t-test, t(27) = 0.5811, P = 0.2849; Fig. 3b). Thus, these results together with our neurochemical findings are in accordance with the hypothesis that consolidation and reconsolidation are similar processes, suggesting that both share a similar or common neural mechanism. Note that the performance during the reactivation in the reconsolidation group showed a significant improvement com-pared with the pre-test (one-sample t-test, t(13) = 2.1937, P = 0.047, d = 0.59, 95% CI: 0.23–30.28), replicating the results of the long-interval condition in study 1.

The results of the behavioural and neurochemical changes provide important implications regarding reactivation and reconsolidation.

a

Reactivation

Control

Day 1 Day 2

0 h 3.5 h

0 h

Baseline

Baseline 3.5 h

b c d

–20

–10

0

10

20

30

40

50

Reactivation Control

Condition

Per

form

ance

impr

ovem

ent (

%)

***

E/I

ratio

cha

nge

(%)

–5

0

5

10

15

20

Baseline 0 h 3.5 h

Session

E/I

ratio

cha

nge

(%)

–5

0

5

10

15

20

Baseline 30 min 3.5 h

Session

Reactivation ConsolidationControl

*

Fig. 2 | Design and results of study 2. a, The design of study 2. There were two groups: the reactivation (n!= !12) and the control (n!= !12) groups. The red-filled boxes represent three test blocks on orientation A, and the red-hatched box represents 16 blocks of training on orientation A. The blue-filled box represents three test blocks on orientation B. The grey boxes represent the MRS measurements. ‘Baseline’, ‘0!h’ and ‘3.5!h’ indicate the MRS measurements performed before, immediately (0!h) after and 3.5!h after the test session on orientation A. b, Performance improvements (mean!± !s.e.m.) in the two groups relative to day!1. A significant main effect of group (***P!= !0.004; see the main text for the ANOVA results) was observed. Dashed grey line at zero mark represents the baseline performance (corresponding to the first three test blocks on orientation A in both groups). c, The mean (± s.e.m.) E/I ratio changes in the early visual areas for the two groups relative to the baseline session. A significant interaction between group and session (*P!= !0.035; see the main text for the ANOVA results) was observed. Dashed grey line at zero mark represents the baseline E/I ratio across both groups. d, The E/I ratio changes (n!= !12; mean!± !s.e.m.) in the early visual areas after encoding, replotted from our previous study24. Dashed grey line at zero mark represents the baseline E/I ratio. Note that the E/I ratio was measured 30!min after training in d, whereas it was measured immediately after reactivation in c. Although these measurement time points were not exactly the same, in both cases, significant interference was observed behaviourally (see ref. 24 and study 1 here). Thus, the E/I ratios at both time points still reflect underlying neurochemical mechanisms in fragile states after training and after reactivation.

NATURE HUMAN BEHAVIOUR | VOL 2 | JULY 2018 | 507–513 | www.nature.com/nathumbehav 509

LETTERS NATURE HUMAN BEHAVIOUR

First, although it has been controversial whether reactivation and reconsolidation of skill learning occurs in humans, our results show that reactivation and reconsolidation indeed occur in humans with VPL, a type of procedural learning.

Second, it has also been a matter of controversy whether the underlying neural mechanisms of reconsolidation after reactiva-tion and consolidation after encoding in VPL are the same. The similarity in the time-course changes in the E/I ratio and behav-ioural performance between reconsolidation and consolidation is in accordance with the hypothesis that these two processes have simi-lar or some common underlying mechanisms.

What do the time-course changes in the E/I ratio in early visual areas in reconsolidation and consolidation processes reflect? A previous study has indicated that the E/I ratio is an index of the degree of plasticity in early visual areas24. Thus, the E/I ratio decreases from high after the end of training or reactivation to baseline levels 3.5 hour later, thereby reflecting that both consoli-dation and reconsolidation are associated with decreases in the degree of plasticity. These time-course changes in the E/I ratio suggest that both consolidation and reconsolidation are driven by homeostasis, in which once enhanced, increased plasticity returns to baseline levels.

Importantly, the E/I ratio does not seem to reflect sensory acti-vation without relating to plasticity. First, a previous study showed that the E/I ratio was not increased after many trials that cause no learning24. Second, the results of the experiment in study 2 of the present paper also showed that the performance of three test blocks, which were not associated with learning, did not increase the E/I ratio. These results together suggest that the E/I ratio is a reliable measure of plasticity.

Our experimental results suggest that reconsolidation occurs during a 3.5-hour interval following reactivation of the trained orientation detection task. However, it remains unclear to what extent these results can be generalized. In addition, these results do not necessarily indicate that the entire process of reconsolidation occurs within the 3.5-hour interval. For example, consolidation occurs during sleep33,34 and during wakefulness. If reconsolidation has a similar mechanism to consolidation, sleep may be involved in reconsolidation. Further studies are needed to fully understand the temporal dynamics of plasticity changes associated with consolida-tion and reconsolidation.

MethodsParticipants. A total of 83 healthy participants (study 1: n = 30, mean age ± s.e.m. = 20.8 ± 0.4, 9 males and 21 females; study 2: n = 24, mean age ± s.e.m. = 21.7 ± 0.5, 11 males and 13 females; study 3: n = 29, mean age ± s.e.m. = 25.7 ± 1.2, 5 males and 24 females) with no use of medication took part in the experiment. All participants had normal or corrected-to-normal vision. They kept regular wake and sleep patterns during experimental days. All participants were informed of the purpose and procedures of the experiment and

gave written informed consent and their demographic information. All procedures were approved by the Institutional Review Board at Brown University.

Gabor stimulus. For the behavioural experiments, participants were presented with Gabor patches with one orientation (contrast = 100%, spatial frequency = 1 cycle per degree, sigma of Gaussian filter = 2.5°, random spatial phase). The centre of the Gabor patches was positioned at the centre of the display. The diameter of the Gabor patches was 4.5°. A noise pattern was generated from a sinusoidal luminance distribution and was superimposed on the Gabor patches at a given signal-to-noise (S/N) ratio. For instance, in the case of a 20% S/N ratio, the noise pattern replaced 80% of the pixels of the Gabor patch.

Orientation detection task. Participants performed a two-interval-forced-choice orientation detection task that was based on a previous study24. In one interval, a Gabor patch was presented with a certain S/N ratio. In the other interval, only a noise pattern (0% S/N ratio) was presented. The interval that included the Gabor patch was determined randomly. Throughout the task, participants were asked to fixate their eyes at the central white bullseye fixation point (diameter = 0.68°). Each trial began with a 500-ms fixation period. After the fixation period, two intervals of stimuli were presented for 50 ms in order, separated by a 300-ms blank period. Participants were asked to determine in which interval (first or second) the Gabor patch appeared by pressing one of two buttons on the keypad. There was no feedback about the correctness of the response.

Threshold measurement. In the orientation detection task, each participant’s threshold S/N ratio in each block was determined by a two-down one-up staircase method in test sessions. This method yielded a 70.7% accuracy rate. There were three blocks per orientation. Within each block, the S/N ratio started with 25% and was further adjusted with a step size of 0.05-log units. Each block was terminated after 10 reversals. Typically, each block consisted of 40 trials and took approximately 1–2 min. We took the geometric mean of the last six reversals within a block as the threshold S/N ratio per block19. The first block served as practice and we took the geometric mean of the threshold S/N ratios across the remaining two blocks as a threshold S/N ratio for its assigned orientation.

Apparatus. We presented visual stimuli on a LCD display (1,024 × 768 resolution, 60-Hz refresh rate) during the orientation detection task in study 1 and on an magnetic resonance imaging (MRI)-compatible LCD display (1,024 × 768 resolution, 60-Hz refresh rate) during MRS experiments in study 2. Unlike studies 1 and 2 in which an LCD monitor was used, in study 3, a CRT display (1,024 × 768 resolution, 60-Hz refresh rate) was used. Gamma correction was applied to the display in each study. All visual stimuli were created using MATLAB and PsychToolbox 3 (ref. 35).

Experimental design for study 1. Experiments were conducted during daytime. There were two groups: the short-interval group (n = 15) and the long-interval group (n = 15). The difference between the groups was the time interval between the offset of reactivation and the onset of the following test. Participants were assigned to either the short-interval or long-interval groups in a counterbalanced manner.

Participants performed the orientation detection task (see above) using two orientations (10° and 70° from the vertical) for two training sessions. Each of the two orientations was randomly assigned to orientation A (the first trained orientation) and orientation B (the second trained orientation) across participants.

The entire behavioural experiment consisted of 3 consecutive days. On day 1, participants in both groups were given a brief test session consisting of three blocks of orientation A. The purpose of the test session was to measure the initial threshold (see above) for orientation A. After the test session, participants were trained on orientation A for 16 blocks.

Day 1

3.5 h

Consolidation

Reconsolidation

a

3.5 hDay 1 Day 2

b

Per

form

ance

impr

ovem

ent (

%) NS

–30

–20

–10

0

10

20

30

Con ReconGroup

Fig. 3 | Design and results of study 3. a, The design of study 3. There were two groups: the consolidation (n!= !15) and the reconsolidation (n!= !14) groups. The red-filled boxes represent three test blocks on orientation A, and the red-hatched boxes indicate 16 blocks of training on orientation A. b, The mean (± s.e.m.) performance change 3.5!h after encoding in the consolidation (Con) group and the mean (± s.e.m.) performance change between reactivation and 3.5!h after reactivation in the reconsolidation (Recon) group. There was no significant (NS) difference in performance changes following the 3.5!h intervals of consolidation and reconsolidation between the groups.

NATURE HUMAN BEHAVIOUR | VOL 2 | JULY 2018 | 507–513 | www.nature.com/nathumbehav510

LETTERSNATURE HUMAN BEHAVIOUR

On day 2, participants in both groups performed a test session (three blocks) of orientation A for the purpose of testing performance improvements from day 1 and also for reactivating VPL of orientation A. In the short-interval group, this test session was immediately followed by another test (< 1 min) and training sessions of orientation B. Participants in the long-interval group performed the test and training sessions of orientation B 3.5 hours after the test session of orientation A.

On day 3, both groups of participants performed a test session of orientation A. By comparing the performance change on orientation A between day 2 and day 3, we examined whether training of orientation B interfered with the reactivated VPL of orientation A.

Experimental design for study 2. There were two groups, the reactivation (n = 12) and the control (n = 12) groups. Participants were assigned to one of the groups in a counterbalanced manner. The only difference between the groups was the procedure on day 1: in the reactivation group, learning on the first trained orientation (orientation A) was expected to occur and consolidate overnight, based on the findings in an earlier study24, whereas learning on orientation A was not predicted for the control group.

Study 2 consisted of 2 consecutive days. In the reactivation group, on day 1, participants performed a brief test session of three blocks on orientation A to measure the initial thresholds (see above). Then, participants completed a training session on orientation A consisting of 16 blocks. Participants performed this behavioural task inside an MRI simulator that looks identical to a real 3 T MR scanner. On day 2, there was a test session of three blocks inside the real 3 T MR machine. It is important to note that this test session served to reactivate previously consolidated VPL of orientation A. There were MRS scans (see below) immediately before, immediately after (< 1 min, corresponding to 0 h in Fig. 2a) and 3.5 hours after the test session.

In the control group, the procedure on day 1 was different from the reactivation group, whereas the rest of the procedures were identical with the reactivation group. On day 1, after the initial three test blocks for orientation A, participants were given another three test blocks for orientation B.

In both groups, there were three MRI sessions to collect MRS data on day 2. The first MRS scan served as the baseline. The second MRS scan was conducted immediately after the test session (0-h scan) and the last scan was conducted 3.5 h after the test session (3.5-h scan). The first and second MRS scans as well as the test session between the MRS scans were conducted sequentially while participants remained inside the scanner.

Experimental design for study 3. There were two groups, the consolidation (n = 15) and the reconsolidation (n = 14) groups. Participants were assigned to one of the groups in a counterbalanced manner.

In the consolidation group, the entire procedure of the experiment was completed within 1 day. Only orientation A was used. Participants from the consolidation group performed a pre-test for the initial threshold measurements that consisted of three blocks. The pre-test was followed by 16 blocks of training. Three and a half hours after the offset of training, participants completed a retest session, which included three test blocks. Performance changes related to consolidation were calculated by the following ratio: ((threshold at pre-test − threshold at retest)/threshold at pre-test) × 100.

In the reconsolidation group, the experiment was conducted over the course of 2 days. Only orientation A was used. On day 1, participants performed a pre-test session, consisting of 3 test blocks, and a training session, consisting of 16 blocks. On day 2, there was a reactivation session, which included 3 test blocks. Three and a half hours after the offset of reactivation, participants completed a retest session, which again consisted of three test blocks. Performance during reactivation and retest was calculated by the following ratio: ((threshold at pre-test − threshold at reactivation or retest)/threshold at pre-test) × 100. Then, performance during reactivation was subtracted from performance during retest to calculate the performance changes due to reconsolidation.

MRI data acquisition. Participants were scanned inside a 3 T MR scanner (Siemens Trio/Prisma) with a 32-channel head coil at the Brown University MRI Research Facility.

First, for anatomical reconstruction, high-resolution T1-weighted MR images were acquired using a multi-echo magnetization-prepared rapid gradient echo (256 slices, voxel size = 1 × 1 × 1 mm, 0-mm slice gap, repetition time (TR) = 2,530 ms, echo time 1 (TE1) = 1.64 ms, TE2 = 3.5 ms, TE3 = 5.36 ms, TE4 = 7.22 ms, flip angle = 7.0°, field of view = 256 mm, bandwidth = 651 Hz per pixel).

Second, based on the anatomical images, a voxel placement for MRS acquisitions for early visual areas was conducted. We positioned a voxel (2 × 2 × 2 cm) manually along the calcarine sulci in the most posterior part of the occipital lobe bilaterally such that the voxel covered early visual areas while minimizing contamination from unnecessary tissues containing lipids. The voxel positioning was carefully replicated during the third MRS acquisition, as the first and second MRS scans were conducted consecutively while participants remained in the scanner. The mean overlap ratio in voxel positioning across scans was > 90%24.

Third, an automatic shimming was performed by a vendor-provided automated shim tool, then later manually for finer adjustments. The mean ( ± s.e.m.) shim value (water linewidth) across three sessions was 14.93 ± 0.22 Hz.

Finally, we measured the concentration of GABA and glutamate from the voxel. The GABA data were obtained using a MEGA-PRESS sequence (TR = 1,500 ms, TE = 68 ms, number of average = 256, scan time = 774 s) with double-banded pulses36–39. Double-banded pulses were utilized to suppress the water signal and to edit the γ -CH2 resonance of GABA at 3 ppm. We subtracted the signals of alternate scans with the selective double-banded pulse applied at 4.7 and 7.5 ppm (‘edit off ’) from those with the selective double-banded pulse applied at 1.9 and 4.7 ppm (‘edit on’) to produce the final spectra. The glutamate data were obtained by the PRESS sequence (TR = 3,000 ms, TE = 30 ms, number of average = 128, scan time = 384 s)40,41. In both GABA and glutamate sequences, a variable pulse power and optimized relaxation delays (VAPOR) technique was used for effective water suppression42. See Supplementary Fig. 2 for exemplary spectra for PRESS and MEGA-PRESS sequences.

In an independent data set (n = 3), the frequency drifts for the MEGA-PRESS sequence were measured24. The MEGA-PRESS sequence was conducted three times for each participant at similar time intervals as in study 2. The mean (± s.e.m.) frequency drifts for the GABA scans were 0.810 ± 0.034 Hz for the first MRS scan, 0.950 ± 0.212 Hz for the second MRS scan and 0.854 ± 0.113 Hz for the third MRS scan. The mean value of within-participant standard deviations was 0.161 Hz24.

Fixation task. During MRS scans, participants performed a fixation task in which they fixated their eyes on the centre of the screen and reported the colour change of the centre dot (0.34° in radius) by pressing a button on a keypad. The purpose of the central task was to keep participants’ fixation at the centre of the screen and the level of attention and vigilance constant across the MRS scans. A fixation dot was placed on a grey disk and the colour of the dot changed unpredictably from white (R, G, B = 255, 255, 255) to faint pink (R, G, B = 255, 255− X, 255− X) and returned to white 1.5 s later. Initially, the colour change X was set to 40 and controlled by a 2-down 1-up staircase method, based on a previous study24 that confirmed that all participants were able to clearly see the colour change with this initial value. If participants pressed the button within 1.5 s after a colour change, this response was regarded as a hit. However, if participants did not press the button within the 1.5-s time interval, it was regarded as a miss. For each scan, we took the geometric mean of the colour change values from the last six trials as a threshold for the degree of colour change.

In the reactivation group, the mean (± s.e.m.) colour change values at the baseline, 0 h and 3.5 h were 16.34 ± 2.86, 15.94 ± 2.53 and 20.33 ± 5.40 during the GABA scan, and 19.16 ± 3.29, 17.04 ± 2.10 and 16.20 ± 2.48 for the glutamate scan, respectively. In the control group, the mean ( ± s.e.m.) colour change values at the baseline, 0 h and 3.5 h were 18.61 ± 1.67, 23.28 ± 2.51 and 26.34 ± 3.85 for the GABA scan, and 19.23 ± 1.20, 22.27 ± 2.48 and 20.48 ± 1.95 for the glutamate scan, respectively.

We tested whether there were any performance differences in the fixation task between groups, scans or across sessions. To this aim, we conducted a three-way mixed ANOVA on the degree of colour change with the factors group (reactivation versus control), scan (GABA versus glutamate; see below) and session (baseline, 0 h and 3.5 h). The results did not indicate any significant main effect of group (F(1,22) = 2.042, P = 0.167), scan (F(1,22) = 0.955, P = 0.339) or session (F(2,44) = 0.917, Huynh–Feldt correction, ε = 0.872, P = 0.396). There was no significant three-way interaction (F(2,44) = 0.003, Huynh–Feldt correction, ε = 0.825, P = 0.993), no significant two-way interactions between session × group (F(2,44) = 1.048, Huynh–Feldt correction, ε = 0.872, P = 0.352) and scan × group (F(1,22) = 0.828, P = 0.373). Although there was a significant session × scan interaction (F(2,44) = 4.164, Huynh–Feldt correction, ε = 0.825, p = 0.03), there was no significant simple main effect of session for GABA (F(2,21) = 2.616, P = 0.097) or for glutamate (F(2,21) = 0.348, P = 0.71), nor a significant simple main effect of scan during any of the sessions (baseline: F(1,22) = 2.492, P = 0.129; 0 h: F(1,22) = 0.003, P = 0.959; 3.5 h: F(1,22) = 3.294, P = 0.083). These results suggest that there were no significant differences in performance on the fixation task across groups, scans or sessions.

MRS data analysis. LC-Model was used in all MRS data analysis43. Note that glutamate and glutamine were separately fitted by the LC-Model and that the concentration of glutamate was used for the calculation of E/I ratio changes. The Cramer–Rao lower bounds (CRLB) show the reliability of quantification of GABA and glutamate. The mean (± s.e.m.) CRLB percentage was 5.486 ± 0.194% for GABA scans across participants and 4.847 ± 0.065% for glutamate scans.

We normalized the amount of GABA and glutamate by using the amount of N-acetylaspartate (NAA), which is a standard reference resonance44 and taken from the glutamate scan following the approach of a previous study24.

We calculated the E/I ratio change at each MRS session by using the equation:

∕ = ∕∕ − ×

⎛

⎝⎜⎜⎜

⎞

⎠⎟⎟⎟⎟

t t tE I change( ) Glu( ) GABA( )Glu(1) GABA(1)

1 100

NATURE HUMAN BEHAVIOUR | VOL 2 | JULY 2018 | 507–513 | www.nature.com/nathumbehav 511

LETTERS NATURE HUMAN BEHAVIOUR

Here, GABA(t) and Glu(t) represent the normalized concentrations of GABA and glutamate by NAA, respectively, at a certain MRS session t (1 = baseline, 2 = 0 h and 3 = 3.5 h).

The NAA concentrations were not affected by group or session. A two-way mixed ANOVA with the factors session (baseline, 0 h and 3.5 h) and group (reactivation versus control) on NAA concentrations showed no significant main effect of session (F(2,44) = 0.864, P = 0.429), no significant main effect of group (F(1,22) = 0.000, P = 0.994) and no interaction between session and group (F(2,44) = 1.169, P = 0.320).

The average linewidth of NAA (± s.e.m) was 8.733 ± 0.165 Hz for GABA scans, 8.781 ± 0.181 Hz for glutamate scans and 8.757 ± 0.122 Hz for all scans combined.

It is important to note that the overall E/I ratio patterns were maintained when GABA and glutamate were normalized to another common control metabolite, such as creatine, owing to the fact that the contribution of the control metabolite is cancelled out in the calculation of E/I ratios.

Exclusion of participants. In addition to the total of 24 participants who participated in study 2, we obtained 3 more participants’ MRS data but excluded their data from all analyses. The first participant exhibited (substantiated by self-report after the measurements) substantial head movements during the scan. The second participant exhibited an extremely high CRLB percentage, such as 999%, and the CRLB percentage is recommended to be < 20%43. The third participant showed a low S/N value in LC-Model fitting. This value seemed smaller than other values and was determined to be an outlier (Grubbs’ test, G = 3.76, P = 9.058 × 10−5) in study 2. In study 3, there were originally 30 participants. However, we excluded one participant because her performance deviated from those of the other subjects on day 2 (Grubbs’ test, G = 2.41, P = 0.0495).

Statistics. Data collection and analyses were not conducted blindly by the investigators with respect to each participant’s group assignment. For all statistical tests, the two-tailed α -level was set to 0.05. The sample size per group was estimated by previous similar experiments on neurochemical changes in the early visual areas using MRS and VPL24. For the majority of analyses, parametric statistical tests (for example, t-tests and ANOVA) were used, following confirmation of the normality of data using the Kolmogorov–Smirnov test. For ANOVAs, we also tested whether the equality of error variances of the dependent variable differed across groups by Levene’s test. Significant differences in error variances between groups were only evident for the comparison of normalized metabolite (GABA and glutamate) concentrations at each time point (0 h and after 3.5 h) between reconsolidation and consolidation. Thus, we used a Mann–Whitney U-test instead of ANOVA. For repeated-measures ANOVA, Mauchly’s test of sphericity was used to test the assumption of sphericity. We applied the Huynh–Feldt correction only when the sphericity was violated and report the estimated epsilon.

Reporting Summary. Further information on experimental design is available in the Nature Research Reporting Summary linked to this article.

Code availability. The computer codes are available from the corresponding author upon request.

Data availability. The data that support the finding of this study are available from the corresponding author upon request.

Received: 13 September 2017; Accepted: 18 May 2018; Published online: 9 July 2018

References 1. Alvarez, P. & Squire, L. R. Memory consolidation and the medial

temporal lobe: a simple network model. Proc. Natl Acad. Sci. USA 91, 7041–7045 (1994).

2. Dudai, Y. The neurobiology of consolidations, or, how stable is the engram? Annu. Rev. Psychol. 55, 51–86 (2004).

3. Dayan, E., Laor-Maayany, R. & Censor, N. Reward disrupts reactivated human skill memory. Sci. Rep. 6, 28270 (2016).

4. Monfils, M. H., Cowansage, K. K., Klann, E. & LeDoux, J. E. Extinction–reconsolidation boundaries: key to persistent attenuation of fear memories. Science 324, 951–955 (2009).

5. Bjorkstrand, J. et al. Disrupting reconsolidation attenuates long-term fear memory in the human amygdala and facilitates approach behavior. Curr. Biol. 26, 2690–2695 (2016).

6. Nader, K., Schafe, G. E. & Le Doux, J. E. Fear memories require protein synthesis in the amygdala for reconsolidation after retrieval. Nature 406, 722–726 (2000).

7. Walker, M. P., Brakefield, T., Hobson, J. A. & Stickgold, R. Dissociable stages of human memory consolidation and reconsolidation. Nature 425, 616–620 (2003).

8. Robertson, E. M. New insights in human memory interference and consolidation. Curr. Biol. 22, R66–R71 (2012).

9. Wood, N. E. et al. Pharmacological blockade of memory reconsolidation in posttraumatic stress disorder: three negative psychophysiological studies. Psychiatry Res. 225, 31–39 (2015).

10. Bos, M. G., Beckers, T. & Kindt, M. Noradrenergic blockade of memory reconsolidation: a failure to reduce conditioned fear responding. Front. Behav. Neurosci. 8, 412 (2014).

11. Hardwicke, T. E., Taqi, M. & Shanks, D. R. Postretrieval new learning does not reliably induce human memory updating via reconsolidation. Proc. Natl Acad. Sci. USA 113, 5206–5211 (2016).

12. Lee, J. L., Everitt, B. J. & Thomas, K. L. Independent cellular processes for hippocampal memory consolidation and reconsolidation. Science 304, 839–843 (2004).

13. Debiec, J., Doyere, V., Nader, K. & Ledoux, J. E. Directly reactivated, but not indirectly reactivated, memories undergo reconsolidation in the amygdala. Proc. Natl Acad. Sci. USA 103, 3428–3433 (2006).

14. Karni, A. & Sagi, D. Where practice makes perfect in texture discrimination: evidence for primary visual cortex plasticity. Proc. Natl Acad. Sci. USA 88, 4966–4970 (1991).

15. Ahissar, M. & Hochstein, S. Task difficulty and the specificity of perceptual learning. Nature 387, 401–406 (1997).

16. Dosher, B. A. & Lu, Z. L. Perceptual learning reflects external noise filtering and internal noise reduction through channel reweighting. Proc. Natl Acad. Sci. USA 95, 13988–13993 (1998).

17. Yu, Q., Zhang, P., Qiu, J. & Fang, F. Perceptual learning of contrast detection in the human lateral geniculate nucleus. Curr. Biol. 26, 3176–3182 (2016).

18. Amar-Halpert, R., Laor-Maayany, R., Nemni, S., Rosenblatt, J. & Censor, N. Memory reactivation improves visual perception. Nat. Neurosci. 20, 1325–1328 2017).

19. Xiao, L. Q. et al. Complete transfer of perceptual learning across retinal locations enabled by double training. Curr. Biol. 18, 1922–1926 2008).

20. Watanabe, T. et al. Greater plasticity in lower-level than higher-level visual motion processing in a passive perceptual learning task. Nat. Neurosci. 5, 1003–1009 2002).

21. Watanabe, T., Nanez, J. E. & Sasaki, Y. Perceptual learning without perception. Nature 413, 844–848 (2001).

22. Seitz, A. R. & Watanabe, T. Psychophysics: is subliminal learning really passive? Nature 422, 36 (2003).

23. Yotsumoto, Y., Watanabe, T. & Sasaki, Y. Different dynamics of performance and brain activation in the time course of perceptual learning. Neuron 57, 827–833 (2008).

24. Shibata, K. et al. Overlearning hyperstabilizes a skill by rapidly making neurochemical processing inhibitory-dominant. Nat. Neurosci. 20, 470–475 (2017).

25. Shibata, K., Watanabe, T., Sasaki, Y. & Kawato, M. Perceptual learning incepted by decoded fMRI neurofeedback without stimulus presentation. Science 334, 1413–1415 (2011).

26. Nikolova, S., Stark, S. M. & Stark, C. E. 3T hippocampal glutamate–glutamine complex reflects verbal memory decline in aging. Neurobiol. Aging 54, 103–111 (2017).

27. Kim, S., Stephenson, M. C., Morris, P. G. & Jackson, S. R. tDCS-induced alterations in GABA concentration within primary motor cortex predict motor learning and motor memory: a 7 T magnetic resonance spectroscopy study. NeuroImage 99, 237–243 (2014).

28. Stagg, C. J., Bachtiar, V. & Johansen-Berg, H. The role of GABA in human motor learning. Curr. Biol. 21, 480–484 (2011).

29. Hensch, T. K. Critical period plasticity in local cortical circuits. Nat. Rev. Neurosci. 6, 877–888 (2005).

30. Stagg, C. J. Magnetic resonance spectroscopy as a tool to study the role of GABA in motor-cortical plasticity. NeuroImage 86, 19–27 (2014).

31. Okubo, Y. et al. Imaging extrasynaptic glutamate dynamics in the brain. Proc. Natl Acad. Sci. USA 107, 6526–6531 (2010).

32. Myers, J. F., Evans, C. J., Kalk, N. J., Edden, R. A. & Lingford-Hughes, A. R. Measurement of GABA using J-difference edited 1H-MRS following modulation of synaptic GABA concentration with tiagabine. Synapse 68, 355–362 (2014).

33. Watanabe, T. & Sasaki, Y. Perceptual learning: toward a comprehensive theory. Annu. Rev. Psychol. 66, 197–221 (2015).

34. Sasaki, Y., Nanez, J. E. & Watanabe, T. Advances in visual perceptual learning and plasticity. Nat. Rev. Neurosci. 11, 53–60 (2010).

35. Brainard, D. H. The Psychophysics Toolbox. Spat. Vis. 10, 433–436 (1997). 36. Hu, Y., Chen, X., Gu, H. & Yang, Y. Resting-state glutamate and GABA

concentrations predict task-induced deactivation in the default mode network. J. Neurosci. 33, 18566–18573 (2013).

37. Mescher, M., Merkle, H., Kirsch, J., Garwood, M. & Gruetter, R. Simultaneous in vivo spectral editing and water suppression. NMR Biomed. 11, 266–272 (1998).

NATURE HUMAN BEHAVIOUR | VOL 2 | JULY 2018 | 507–513 | www.nature.com/nathumbehav512

LETTERSNATURE HUMAN BEHAVIOUR

38. Rothman, D. L., Behar, K. L., Hetherington, H. P. & Shulman, R. G. Homonuclear 1H double-resonance difference spectroscopy of the rat brain in vivo. Proc. Natl Acad. Sci. USA 81, 6330–6334 (1984).

39. Robertson, C. E., Ratai, E. M. & Kanwisher, N. Reduced GABAergic action in the autistic brain. Curr. Biol. 26, 80–85 (2016).

40. Hancu, I. Optimized glutamate detection at 3T. J. Magn. Reson. Imaging 30, 1155–1162 (2009).

41. Mullins, P. G., Chen, H., Xu, J., Caprihan, A. & Gasparovic, C. Comparative reliability of proton spectroscopy techniques designed to improve detection of J-coupled metabolites. Magn. Reson. Med. 60, 964–969 (2008).

42. Tkac, I., Starcuk, Z., Choi, I. Y. & Gruetter, R. In vivo 1H NMR spectroscopy of rat brain at 1 ms echo time. Magn. Reson. Med. 41, 649–656 (1999).

43. Provencher, S. W. Estimation of metabolite concentrations from localized in vivo proton NMR spectra. Magn. Reson. Med. 30, 672–679 (1993).

44. Provencher, S. W. Automatic quantitation of localized in vivo 1H spectra with LCModel. NMR Biomed. 14, 260–264 (2001).

AcknowledgementsThis work was supported by the NIH (R01EY019466), the NSF (BCS 1539717) and the JSPS KAKENHI grant number 17H04789. The funding agencies had no role in the

conceptualization, design, data collection, analysis, decision to publish or preparation of the manuscript.

Author contributionsThis work was conceived by J.W.B., S.M.F., T.W. and Y.S. J.W.B., K.S., S.M.F. and E.G.W. collected and analysed the data. J.W.B., K.S., S.M.F., E.G.W., M.W.G., T.W. and Y.S. wrote the manuscript.

Competing interestsThe authors declare no competing interests.

Additional informationSupplementary information is available for this paper at https://doi.org/10.1038/s41562-018-0366-8.Reprints and permissions information is available at www.nature.com/reprints.Correspondence and requests for materials should be addressed to T.W.Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

NATURE HUMAN BEHAVIOUR | VOL 2 | JULY 2018 | 507–513 | www.nature.com/nathumbehav 513